beaunix/aegis-geo-mind-qwen2.5-7b
Aegis-Geo-Mind-Qwen2.5-7B is a 7.6 billion parameter Qwen2.5-7B-Instruct model fine-tuned by beaunix using QLoRA. This model specializes in geology and Earth-science, with a focus on petroleum geology concepts like sedimentology, reservoir stratigraphy, well logging, and seismic interpretation. It is designed as an educational and research assistant, producing fluent, domain-appropriate explanations using correct terminology across various geological disciplines. The model excels at expert geological reasoning style but requires factual verification due to potential inaccuracies in specific numeric data.
Loading preview...
Aegis-Geo-Mind (Qwen2.5-7B, QLoRA) Overview
This model, developed by beaunix, is a Qwen2.5-7B-Instruct variant fine-tuned with QLoRA (4-bit, rank 16) on a specialized geoscience corpus. It focuses on geology and Earth-science, particularly petroleum geology, encompassing sedimentology of petroliferous basins, reservoir stratigraphy, well logging, and seismic interpretation.
Key Capabilities
- Domain-specific explanations: Generates fluent, expert-style explanations using correct terminology in structural geology, stratigraphy, sedimentology, petroleum systems, and well-log interpretation.
- Educational and research assistant: Intended to support learning and research in geology and petroleum geology concepts.
Training Details
The model was fine-tuned on a sequence length of 1024 tokens using a curated corpus. This corpus combines filtered geoscience QA (from a public GeoSignal subset) with general and petroleum geology QA pairs, all deduplicated and quality-filtered.
Important Limitations
As a supervised fine-tune without retrieval augmentation, Aegis-Geo-Mind reliably reproduces the style and structure of expert geological reasoning. However, it can state specific facts, numbers, and ratios incorrectly with high confidence. Known weak areas include:
- Kerogen type classification (H/C and O/C ratios)
- Numeric ranges (oil-window temperatures, burial depths, maturity cutoffs)
- Occasional invented specific values
Users must verify all factual or numeric outputs against authoritative sources. This model is not for operational decisions in exploration, drilling, or any geological operations. A RAG-enabled version is planned to address these factual gaps.